Cooperation in WTO’s Tariff Waters∗

Alessandro Nicita†

Marcelo Olarreaga‡

Peri Silva§

May 2013

Abstract

Many members of the World Trade Organization apply tariffs well below the negotiatedtariff bounds, a phenomenon known as “tariff water”. In principle, there is no legal constraintto set these tariffs to exploit the importing country’s market power in international markets.We found that this is the case, but only in the presence of deep tariff water. Indeed, if thedifference between applied and bound tariffs is below 20 percent, tariffs of WTO memberscannot be explained by their market power, suggesting that cooperation in the WTO goesbeyond observed tariff bounds. Cooperation within WTO tariff waters can be explained bythe fear of retaliation from trading partners, in particular when these have considerable marketpower and tariff water in their schedules.JEL classification numbers: F1Keywords: Export supply elasticities, WTO cooperation, tariff water

∗We are grateful to Sajal Lahiri, Jaime de Melo, Christoph Moser, Patrick Low, Subhash Sharmaand and seminarparticipants at ETH Zurich, Southern Illinois University and Sussex University for helpful comments and discussions.The views expressed here are those of the authors and do not necessarily reflect those of the institutions to whichthey are affiliated.†UNCTAD, email: alessandro.nicita@unctad.org‡University of Geneva and CEPR, email: marcelo.olarreaga@unige.ch.§Kansas State University and Centro Studi Luca d’Agliano, email: pdasilva@k-state.edu.

1 Introduction

As predicted by trade policy models, there is strong empirical evidence suggesting non mem-

bers of the World Trade Organization (WTO) set import tariffs non-cooperatively to exploit their

market power in international markets (Broda, Limao and Weinstein, 2008). The rationale is that

a tariff reduces import demand, which in the absence of a perfectly elastic export supply reduces

import prices, and therefore improves the terms-of-trade of the importing country. By definition,

this also reduces the terms-of-trade of the exporting country, therefore creating the potential for

cooperation. Recent evidence suggests that WTO negotiations do facilitate cooperation in tariff set-

ting by allowing recently acceding WTO members to internalize these terms-of-trade externalities,

resulting in tariff schedules that do not reflect their market power (Bagwell and Staiger, 2011).

However, even within the WTO legal framework, many member countries maintain a lot of

flexibility that could potentially allow them to set non-cooperative tariffs to exploit their market

power. This flexibility is provided by the presence of what is called “tariff water”, or “binding

overhang” (i.e. the difference between WTO legally binding tariffs and applied tariffs). In some

countries, these legally binding tariffs are multiples of the applied tariffs, providing them with a lot

of policy space. In principle, the absence of tariff water indicates cooperation in tariff setting as

the importing country is bound by its commitments to other trading partners. On the other hand,

the presence of tariff water provides WTO members with the opportunity to set WTO-consistent

tariffs at non-cooperative levels.1

In this paper we explore whether cooperation in tariff settings extends beyond the setting

of binding tariffs to include the degree of discretion provided by tariff water. In particular, we

investigate whether in the presence of tariff water the applied tariffs reflect WTO members’ market

power. Cooperation beyond bound tariffs may result from fear of retaliation by trading partners

which can also legally raise their tariffs within tariff waters, or more general forms of retaliation.

To guide our empirical work, we consider a two-country model in which tariffs are driven by

terms-of-trade rationales, as well as political economy forces. In our theoretical setting, countries

can set tariffs cooperatively depending on the trade-off between benefits and costs of cooperation.

1Amador and Bagwell (2012) explain the possible presence of tariff water based on a model where uncertainty andprivate information are present. In their model, there are no costs in negotiating an agreement across countries. Onthe contrary, Horn, Maggi and Staiger (2010) suggest that tariff water may exist in the presence of uncertainty andcontract costs.

We assume that in the absence of cooperation, an exogenous tariff bound is set for the importing

country allowing it to implement a non-cooperative tariff policy. Conversely, in the presence of

cooperation, the negotiated tariff maximizes the joint political function of the two countries. In

the former case, tariff water is present since the exogenous bound is assumed to be high enough to

allow for non-cooperative behavior. In the latter case, water is not present presuming that perfect

information is present in this framework. This dichotomy in the treatment of importing countries

seems to fit well with the different manners in which developed and developing countries have so

far participated in multilateral agreements as discussed in Croome (1995) and in Hoekman and

Kostechi (2009).

The theoretical predictions of our model show that in the absence of cooperation the import-

ing country sets tariffs that exploit its market power. However, in the presence of cooperation,

the importing country’s tariffs are inversely related to its market power, as a higher tariff would

significantly diminish the profits accrued by the politically influential producers in the exporting

country.

We empirically test the predictions by examining whether WTO members are more likely to set

non-cooperative tariffs when tariff water makes this legally possible. Our empirical specification

explains applied tariffs with the degree of market power enjoyed by the importer (i.e., the inverse

of the export supply elasticity of the rest-of-the-world), as well as its interaction with a measure of

the importer’s tariff water.

This framework requires estimates of rest-of-the-world export supply elasticities. The only

available estimates are for non-WTO members (Broda et al. 2008). In order to provide estimates

for all WTO members, we build on Kee et al. (2008) adaptation of Kohli’s (1991) revenue function

approach to the estimation of trade elasticities. We basically estimate the revenue function of

the rest-of-the-world for each WTO member, which is a function of the rest-of-the-world factor

endowments and the price they face in the WTO member’s import market. The price parameter

of the revenue function of the rest-of-the-world can then be used to estimate the export supply

elasticity of the rest-of-the-world in the WTO member’s market as in Kee et al. (2008).

We estimated more than 260,000 export supply elasticities of the rest-of-the-world faced by

119 importing countries at the six-digit level of the Harmonized System (HS) classification.2 The

2As discussed in the empirical Section, we present our results using the European Union members as one countryfor trade policy purposes. We also estimate results using data on individual European countries and obtain similar

2

median of the inverse of the export supply elasticity is 0.044, suggesting that if countries were to

set tariffs non-cooperatively, the median tariff in the world would be around 4.4 percent. This is

smaller than the 5 percent median tariff we observe in our sample. Because part of the terms-of-

trade rationale vanishes through cooperation in trade agreements, other forces than terms-of-trade

are needed to explain the tariff levels observed.

With the help of these estimates we found evidence that tariffs are set non-cooperatively in

the presence of substantial tariff water, i.e., when the difference between bound and applied tariffs

is above 20 percentage points. Below this threshold the correlation between market power and

applied tariffs tends to be negative, suggesting that there is cooperation in the presence of moderate

amounts of tariff water.

The presence of cooperation within tariff waters calls for an explanation. A potential candidate

is that member’s fear of retaliation by their trading partners with significant amount of market

power and tariff water may prevent them from setting tariffs non-cooperatively within WTO’s tariff

waters. Fear of retaliation has been shown to be an important determinant of trade policy. Indeed,

Blonigen and Bown (2003) show that retaliation threats makes less likely to observe andtidumping

measures by the United States. Similarly, Bown (2008) show that fear of retaliation makes WTO’s

dispute settlement defendants more likely to comply with their WTO commitments.

After building a measure of fear of retaliation that captures trading partner’s market power and

the scope for tariff increases within their tariff bindings, we found that non-cooperative tariff setting

was observed only when importing countries had relatively low fear of retaliation by their trading

partners. In other words, non-cooperative behavior within WTO tariff waters is only observed for

those members which have little to lose from retaliation by their trading partners.

The remainder of the paper is organized as follows. Section 2 provides the theoretical framework

and describes our empirical strategy. Section 3 focuses on the estimation of rest-of-the-world export

supply elasticities faced by each importer. Section 4 presents the empirical results regarding the

extent of cooperation in tariff setting in the WTO’s tariff waters. Section 5 concludes.

results. Notice that using data for individual European Union members raises the number of estimated elasticities tomore than 300,000 and the total number of countries in our sample raises to 131. The results using data for individualEuropean countries are available upon request.

3

2 Optimal tariffs and the WTO

In the absence of cooperation in tariff setting in the WTO, tariffs could be set non-cooperatively

and reflect both market power and domestic political economy rationales as in Grossman and

Helpman (1995) and Bagwell and Staiger (1999). If cooperation takes place then the market power

rationale vanishes as it only reflects inefficient transfers from the exporting country to the importing

country that are internalized through cooperation, and that’s why we shouldn’t expect negotiated

tariffs to reflect the importing country’s market power. However, the presence of tariff overhang or

tariff water opens the door for non-cooperative tariff setting among WTO members. In order to

explore this possibility we first provide an analytical setup where applied tariffs can be set below

tariff bindings due to the presence of negotiation costs. We then provide an empirical strategy to

examine whether tariffs are set non-cooperatively in the presence of tariff water, i.e., reflecting the

importer’s market power.

2.1 Theoretical predictions

Assume a two-country world where cooperation in tariff setting prevails as long as negotiation

costs are lower than the gains from cooperation. This framework only considers import tariffs and

uncertainty is not present. Consequently, in the presence of cooperation, the negotiated applied

tariff matches the bound tariff and tariff water is not present. In the absence of cooperation,

however, we consider that each country uses an exogenous bound tariff assumed to be much higher

than the applied tariff, i.e., tariff water is present, which allows the importing country to set

non-cooperative tariffs. Thus, all tariffs will be bound in the agreement but only some will be

negotiated.

This setting seems to fit well with the manner in which developing countries participated so

far in multilateral agreements. As described in Croome (1995), during the Uruguay Round an

Australian proposal was adopted to ensure that most countries would bound their tariffs by allowing

each member to follow its own approach to tariff binding. This led many developing countries, in

particular the smaller and poorer countries, to bound almost all of their previously unbound tariffs

at arbitrarily high levels.3

3For example: 19 of the 36 least developed countries at the time, bound their tariffs at levels above 100 percent,whereas their applied average tariffs were close to 10 percent. The binding levels were also taken quite arbitrarily.

4

On the other hand, it is clear that the United States, the European Union, and Japan play a

prominent role in negotiating tariffs under the WTO and the available data (see Table 1) indicate

that they have very low applied and bound tariffs, in which the presence of water is almost nonex-

istent, except in the case of a handful of sensitive products. Our model captures this dichotomy.

We consider a home country and a foreign country where foreign country’s variables are iden-

tified by superscript “?”. These countries trade three goods labeled 0, 1 and 2, where good 0

represents a numeraire good that is freely traded. Consumer preferences are the same across coun-

tries and are described by the following additive quasilinear utility function:

U (c0, c1, c2) = c0 + u1 (c1) + u2 (c2) (1)

which describes the preference structure in the home country while a similar expression describes

the preference structure in the foreign country. We assume that sub-utility functions are increasing

on consumption and concave, i.e. u′i (.) > 0 and u

′′i (.) < 0.

On the production side, we assume that the numeraire good is produced using labor under

constant returns to scale, which keeps the wage rate constant regardless of the trade policy imposed

on imports of goods 1 and 2. Moreover, we assume that goods 1 and 2 are produced using labor and

a specific factor needed to produce each good using a constant return to scale technology. Perfect

competition prevails. Thus, the assumptions on the supply side and on the demand side of the

model allow us to conclude that the market equilibrium for good 1 is not affected by the market

equilibrium for good 2.4

We consider that the home country imports good 1 while it exports good 2. This implies

x1 (p) < x?1(p), where x1 and x?1 are the supply of good one in the home and foreign country,

respectively. The reverse happens for good 2. As a result, a tariff on good 1 (2) may be imposed

by country 1 (2) as we only consider tariffs and disregard export-related trade instruments. The

According to interviews with Mauritanian participants in the final Ministerial meeting of the Uruguay Round inMarrakech, their delegation was briefed by the GATT secretariat’s staff in a meeting that lasted a couple of hours ina hotel room in Marrakech. They went over the last eight years of negotiations in Geneva, where Mauritania did nothave a negotiating team, before making a decision on the level at which agriculture and manufacturing tariffs wouldbe bound. More importantly, while most developed countries had locked in their offers before the Marrakech meetingthat concluded the Uruguay Round, many developing countries were still drafting their offers during the Marrakechmeeting, and least-developed-countries had an extra-year to submit their goods and services tariff schedules. Thus,this made bilateral negotiations impossible, and it is therefore not surprising that today many developing countrieshave very large levels of water in their tariff schedules.

4This rules out counterlobbying by exporters within the same country as in Krishna et al. 2012.

5

relationship between the price in the home and in the foreign countries are then described by

p1 = p?1 + t1 and p?2 = p2 + t2. The cost of negotiating each tariff between these two countries is

described by the parameter α which is assumed to be positive. If negotiation costs are high relative

to the benefits of negotiation then the importing country imposes a non-cooperative tariff.

We consider that the home country’s government objective function G (p1, p2) is defined by a

weighted average between profits and social welfare. In this case, parameter β > 0 describes the

extra-weight given to profits relative to consumer surplus and tariff revenue in this government’s

objective function. A similar approach applies to the foreign country’s government where the extra

weight to profits is captured by parameter β?. Then, the home country’s government objective

function is described, with the assistance of expression (1), by the following expression:

G (p1, p2) = u1 (d1 (p1))− p1d1 (p1) + u2 (d2 (p2))− p2d2 (p2) (2)

+t1m1 (p1) + (1 + β) [π1 (p1) + π2 (p2)]

where di is the demand for good i, m1 = d1 − x1 stands for imports of good 1 and π1 stands for

home firms’ profits in sector 1.

The choice of assumptions on the supply and on the demand side, along with separate costs

to negotiate each tariff, allows us to consider the choice of whether to negotiate tariffs on goods 1

and 2 independently. For tractability, we then focus our attention on the decision to negotiate a

tariff for good 1, but a similar analysis applies for good 2. We first investigate the tariff for good

1 that emerges with and without negotiation between the countries. Later, we use the equilibrium

tariffs under the two scenarios to consider the role played by market power and political influence

in determining the benefits of negotiation.

If countries do not cooperate, then the tariff imposed on imports of good 1 by the home country

maximizes its government objective’s function. Total differentiation of expression (2) yields the

first order condition for this country’s government with respect to the choice of tariff on good 1 as

follows,

6

dG

dt1= −d1

[dp?1dt1

+ 1

]+m1 + t1m

′1

[dp?1dt1

+ 1

](3)

+ (1 + β)x1

[dp?1dt1

+ 1

]

where we do not express the demand, supply and import as functions of the domestic price to save

on notation. Expression (3) can be rearranged in the following manner:

dG

dt1= −m1

dp?1dt1

+ t1m′1

dp1

dt1+ βx1

dp1

dt1(4)

where we note that dp1

dt1=

dp?1dt1

+ 1. We can solve for the non-cooperative tariff by setting expression

(4) equal to zero. As usual, we can use the market clearing condition to solve for the non-cooperative

tariff using (4) and express the non-cooperative tariff as a function of the importing country’s market

power. Since imports equal exports we can express the marketing clearing condition as follows:

m1 (p1) +m?1 (p?1) = 0 (5)

and total differentiation of the market clearing conditions yields

m′1dp1

dt1= −m?′

1

dp?1dt1

(6)

We can apply relationship (6) to solve for the non-cooperative tariff using (4) to obtain:

tN1 =βz1p1

e1+p?1e?1

(7)

where tN1 is the non-cooperative optimal tariff, z1 stands for the inverse of the import penetration

ratio, e1 represents the import demand elasticity, and e?1 stands for the export supply elasticity

faced by the importing country. Expression (7) displays the usual two motives for deviations from

free trade without cooperation across countries under perfect competition. The political economy

motive is represented by the first term on the right hand side of (7) while the market power motive,

also known as the terms-of-trade motivation, is described in the second term on the right hand side.

As Bagwell and Staiger (1999) explain in detail, the latter motivation corresponds to a negative

7

externality of the importing country’s trade policy on the exporting country. Negotiations between

countries should internalize this motivation by design while respecting the political economy forces

in each negotiating party.

We can now investigate the equilibrium tariff on good 1 that emerges when the two countries

cooperate. We adopt the usual assumption that negotiated tariffs maximize the sum of the gov-

ernments’ political functions.5 In this case, we represent the sum of the political functions by the

global political function, which is represented by Gw = G+G?.6 Focusing on the equilibrium tariff

for good 1, we can totally differentiate Gw to obtain the following expression:

dGw

dt1= −d1

[dp?1dt1

+ 1

]+m1 + t1m

′1

[dp?1dt1

+ 1

](8)

+ (1 + β)x1

[dp?1dt1

+ 1

]−d?1

dp?1dt1

+ (1 + β?)x?1dp?1dt1

where the first and second lines can be found in expression (3) and the third line comes from

calculating dG?

dt1. Expression (8) can be rearranged to yield the following expression:

dGw

dt1= t1m

′1

dp1

dt1+ βx1

dp1

dt1+ β?x?1

dp?1dt1

(9)

where it is clear that the political economy forces in each country are driving forces in determining

the negotiated tariff. The equilibrium cooperative tariff can be calculated by setting expression (9)

to zero, and with assistance of expression (6), we can rearrange to obtain:

tC1 =βz1p1

e1− β?z?1p

?1

e?1(10)

where (10) is the optimal cooperative tariff, z?1 is the inverse of the export penetration ratio in the

foreign country. It is clear from expression (10) that a cooperative tariff differs from zero due to

the political forces present in each negotiating party. Otherwise, free trade would prevail. Notice

5This follows other papers in the literature such as Bagwell and Staiger (1999), Horn, Maggi and Staiger (2010),and Beshkar, Bond, and Rho (2012), among others.

6The usual rationale for focusing on the joint political payoff is the presence of similar countries in economic andpolitical power or the presence of cross-country transfers. We follow suit in line with the literature.

8

that politically important exporters (β? > 0) influence the cooperative tariff in a very intuitive

way. If the importing country market power is high (low e?1/p?1) then the equilibrium tariff under

cooperation tends to be lower, as a high tariff would cause a significant decrease in the exporting

country’s price, which obviously has a negative effect on the politically influential producers in the

foreign economy. Expressions (7) and (10) allow us to elaborate the following prediction:

Prediction 1 The higher the market power of the importing country, the lower (higher) the equi-

librium tariff when countries (do not) cooperate.

As discussed above, cooperation over tariffs leads to the internalization of the terms of trade

motivation for protection and this seems to be the key contribution of negotiations between coun-

tries. However, cooperation has a price as identified by the parameter α. Thus, we assume that

countries cooperate if the increase in value of the global political function (Gw(tC1)−Gw

(tN1)) is

greater than α. We are looking for the conditions under which the gains from negotiations are

high enough to compensate the costs. Putting it in a formal way, the gains from cooperation are

described by the following expression:

Gw(tC1)−Gw

(tN1)

= −∫ tN1

tC1

dGW , tN1 > tC1 (11)

Following Horn, Maggi and Staiger (2010), we investigate the sufficient condition for obtaining

large gains from cooperation as described by expression (11). By definition, the function Gw is

concave anddGw(tC1 )

dt1= 0 since the cooperative tariff maximizes the global political function. Thus,

a sufficient condition for large gains from cooperation is to have∣∣∣dGw(tN1 )

dt1

∣∣∣ large but this boils down

to have∣∣∣dG?(tN1 )

dt1

∣∣∣ large sincedG(tN1 )

dt1= 0 by definition of the non-cooperative solution. Using the

definition of the foreign country’s objective function we can obtain:

∣∣∣∣dG?(tN1 )

dt1

∣∣∣∣ = (d?1 − x?1)dp?1dt1− β?x?1

dp?1dt1

(12)

Expression (12) can be rearranged with the assistance of expression (6) to yield the following

sufficient condition:

∣∣∣∣dG?(tN1 )

dt1

∣∣∣∣ =(m1 + β?x?1)m

′1(

m′1 +m?′

1

) (13)

9

which can be rewritten to display the relevant elasticities as follows,

∣∣∣∣dG?(tN1 )

dt1

∣∣∣∣ =(m1 + β?x?1)(

1 +e?1p?1

p1

e1

) (14)

We can relate expression (14) to the discussion above about the equilibrium tariffs. This suffi-

cient condition indicates that countries are more likely to cooperate if the importing country has

significant market power (low e?1/p?1), a tariff creates significant distortions in the importing country

(high e1/p1), exporters are politically influential (high β?), and countries trade a great deal with

each other (high m1). If these conditions apply then countries cooperate and tariff water is not

present since the bound and applied tariff are described by the cooperative tariff (10). Otherwise,

countries do not cooperate, water is present, and Prediction 1 suggests that tariffs reflect the market

power of the importing country. This is summarized in the following prediction:

Prediction 2 If gains from cooperation described by expression (14) are relatively low (large) com-

pared to negotiation costs, then tariff water is (not) present and tariffs are positively (inversely)

related to market power.

2.2 Empirical strategy

In order to empirically test the two predictions developed in the previous section, we will use tariff

data for 113 WTO members at the six-digit of the HS classification,7 and investigate the extent to

which the importer’s market power (the inverse of the export supply elasticity of the rest-of-the-

world) can explain the variation in tariffs, in particular in the presence of tariff water. The two

predictions are that in a world where countries can cooperate in tariff setting: i) the importer’s

market power should lead to lower tariffs as other WTO members have stronger incentives to

negotiate tariff cuts, but ii) in the presence of tariff water, the importer’s market power leads to

higher tariffs. In order to test these two predictions we start with the following specification:

tp,c,t = α1 ×1

e?p,c+ α2 ×Wp,c,t + α3 ×

1

e?p,c×Wp,c,t + αp + αc,2HS,t + µp,c,t (15)

where tp,c,t is the applied tariff in product p (defined at the six-digit of the HS classification) in

7For a list of countries, see Table 1.

10

country c at time t, W captures tariff water which is measured as the difference between bound and

applied tariffs, αp is a product fixed effect defined at the six-digit level of the HS classification, and

αc,2HS,t is a two-digit HS fixed effect that varies by country and by year, which serves as control

for political economy determinants of tariffs, such as firm concentration, capital/labor intensity

etc.8 Our predictions will therefore be identified using the variation across HS six-digit tariff lines

within HS two-digit aggregates for each country and year, while controlling for HS six-digit common

effects. Our first prediction suggests that α1 < 0, whereas the second prediction implies α3 > 0.

There are several issues with the estimation of (15). First, export supply elasticities of the

rest-of-the-world are measured with a lot of noise as suggested by Broda et al. (2008).9 We follow

their strategy and use as an alternative the log of 1/e?, as well as dummy variables that split the

sample into high, medium and low levels of market power across all countries, products and time.

This last alternative also fits better our analytical setup since it suggests that above a certain level

of market power, tariffs tend to be lower than below the threshold, without necessarily implying a

variation with the levels of market power below or above the threshold.

The second issue has to do with the endogeneity of our measure of tariff water and market

power. We solve the endogeneity of tariff water by instrumenting it with what Foletti et al. (2011)

labeled as water vapour:

Water Vapourp,c,t = max{

0, tbp,c − tprp,c,t

}(16)

where tbp,c stands for the bound tariff and tprp,c,t for the prohibitive tariff. So water vapour is

tariff water above the prohibitive tariff.10 Arguably, this satisfies the exclusion and the inclusion

restrictions as the level of the applied tariff should not depend on how much water vapour exists,

and by construction, water vapour is correlated with tariff water as it is part of it.

To construct water vapour, we need a measure of prohibitive tariffs for every tariff line. These

are not observables, but we use the approximation in Foletti et al. (2011) which using import

demand elasticities calculates the prohibitive tariff as the one that will lead to zero imports using a

8Ideally, we would like to have these types of controls varying at the six-digit level of the HS classification, butsuch data does not exist across countries, so a good compromise is to use fixed effects at the two-digit level of the HSclassification.

9We also do not have estimates that vary across time and therefore the only variation in these elasticities is acrossproducts and countries.

10Notice that tariff bounds do not vary by time given that they were the outcome of the Uruguay round negotiations.

11

linear approximation around the observed level of imports. The prohibitive tariff is then given by:

tprp,c,t = tp,c,t +(1 + tp,c,t)

emp,c(17)

where emp,c represents the import demand elasticity which varies by country and by product. Table

1 provides summary statistics by country of tariff water and water vapour, as well as applied tariffs

and bound tariffs.

The endogeneity of market power is addressed by using a bit of theory. Olarreaga et al. (1999)

show that two determinants of the export supply elasticity of the rest-of-the-world are an average

of the export supply elasticity across all countries measured from the exporters point of view and

an average of the import demand elasticities across all countries in the rest-of-the-world.11 We

have estimates of import demand elasticities at the six-digit of the HS classification from Kee et

al. (2008), and we adapt their methodology to estimate export supply elasticities for each country

in our sample at the six-digit of the HS classification. The methodology employed to measure the

export supply elasticities of the rest-of-the-world from the point of view of the importers is discussed

in Section 3. We then take averages of these elasticities and use them as instruments for market

power (the inverse of the export supply elasticity of the rest-of-the-world from the point of view of

the importer). Below, we provide more details on this issue. In principle these two averages satisfy

the exclusion restriction. We instrument the interaction term with the interaction of these averages

with water vapour, which serves as an instrument for water. We perform over-identification and

weak instrumental variables’ tests to check the validity of our instruments.

3 Estimating export supply elasticities of the rest-of-the-world

We start by describing our adaptation of the methodology used in Kee et al.’s (2008) to estimate

export supply elasticities of the rest-of-the-world faced by each importing country (e?nn). We then

11 For a given product, let us define world export supply as xw =∑

c xc (the sum of each country’s export supply).The rest-of-the-world export supply faced by country i is then given by xi = xw −

∑c6=i mc where mc are imports of

country c. Differentiate both sides by the world price p and multiply by p/xw, and rearrange to obtain:

e?i =1

mi/xw

ex? +∑c6=i

emcmc

xw

where ex? is the export supply of the world, and ec is the absolute value of the import demand elasticity of countryc.

12

discuss the adaptation of their methodology to estimate export supply elasticities of each exporting

country at the six-digit level of the HS classification that will be used jointly with the estimates in

Kee et al. (2008) to instrument the export supply elasticities of the rest-of-the-world faced by each

importer. We then describe the data used to estimate the elasticities, provide some descriptive

statistics of these estimates, and also provide further external tests to show that our estimates are

in line with basic economic intuition.

3.1 Estimating rest-of-the-world export supply elasticities

In this section, we describe the methodology employed to estimate rest-of-the-world supply

elasticities faced by each importer. They correspond to our measure of market power in international

markets and capture the ability of countries in changing their terms of trade by using trade policy

instruments, for instance. The empirical model is based on Kee et al. (2008) adaptation of Kohli’s

(1991) GDP function approach for the estimation of trade elasticities at the tariff line level. Kee et

al. (2008) provide estimates of import demand elasticities at the six-digit HS level, whereas here

our focus is the export supply of the rest-of-the-world, so we need to model the GDP function of

the rest-of-the-world for each importing country.

We assume that the GDP function is common across all countries up to a constant term that

accounts for productivity differences. The GDP function of each country, denoted Gt(pt, vt

), is a

function of prices and endowments. Without loss of generality, we assume that this GDP function

has a flexible translog functional form, where n and k index goods, and m and l index factor

endowments, as follows:

lnGt(pt, vt

)= at00 +

N∑n=1

at0n ln ptn +1

2

N∑n=1

N∑k=1

ank ln ptn ln ptk

+M∑

m=1

bt0m ln vtm +1

2

M∑m=1

M∑l=1

btml ln vtm ln vtl

+

N∑n=1

M∑m=1

cnm ln ptn ln vtm, (18)

where all the translog parameters a, b and cs when indexed by t allow for changes over time.12

12We assume some parameters to be time invariant so that we can estimate them using the variation over time.

13

We also impose the necessary restrictions so that the GDP function satisfies the homogeneity and

symmetry properties of a GDP function.13 For each country c we can then construct the GDP

function of the rest-of-the-world by summing the GDP functions of each country given by (18).

Then, taking the derivative of lnGt(pt, vt

)with respect to ln ptn, and summing across each country

c in the rest-of-the-world, we obtain the equilibrium share of exported good n in rest-of-the-world’s

GDP at period t,14

stn(pt, vt

)≡

ptnqtn

(pt, vt

)Gt (pt, vt)

= (Cw − 1)at0n + (Cw − 1)

N∑k=1

ank ln ptk +

M∑m=1

cnm

Cw−1∑c=1

(ln vtm

)c

= (Cw − 1)

at0n + ann ln ptn + ank∑k 6=n

ln ptk

+

M∑m=1

cnm

Cw−1∑c=1

(ln vtm

)c

(19)

where stn is the share of export good n in rest-of-the-world GDP, Cw is the total number of countries

in the world, and∑Cw−1

c=1

(ln vtm

)c

is the sum of the log of factor endowment m across all countries

in the rest-of-the-world.

The rest-of-the-world export supply elasticity of good n is then given by:15

e?nn ≡∂qtn

(pt, vt

)∂ptn

ptnqtn

=(Cw − 1)ann

stn+ stn − 1 > 0 (20)

Thus we can calculate the export supply elasticities once ann is properly estimated. Note

that the size of the export supply elasticities, e?nn positively depends on the size of ann which

captures the changes in the share of good n in each country’s GDP when the price of good n

increases.

With data on export shares, unit values and factor endowments, equation (19) is the basis of

13More specifically:

N∑n=1

at0n = 1,

N∑k=1

ank =

N∑n=1

cnm = 0, ank = akn, ∀n, k = 1, ..., N, ∀m = 1, ...,M.

And:N∑

n=1

bt0n = 1,

N∑k=1

btnk =

M∑m=1

cnm = 0, btnk = btkn, ∀n, k = 1, ..., N, ∀m = 1, ...,M.

14This assumes that goods exported by the rest-of-the-world are differentiated by destination, and the price ofgoods to other destinations are included in the second term of the right-hand-side on the top line of (19).

15Cross-price elasticities of export supply are given by:εtnk ≡∂qtn(pt,vt)

∂ptk

ptkqtn

=atnkstn

+ stk, ∀n 6= k.

14

our estimation of export elasticities. There are, however, several problems with the estimation of

ann using (19). First, there are literally thousands of goods traded among the countries in any

given year. Moreover, there is also a large number of non-traded commodities that compete for

scarce factor endowments and contribute to GDP in each country. Thus, we do not have enough

degrees of freedom to estimate all anks.

We follow Kee et al. (2008) to solve this problem by transforming the N -good economy problem

into a collection of N sets of two-good economies. This is done by constructing for each n exported

good a price index of the remaining of the economy (including imported and non-traded goods).

For this we use information on GDP deflators, a price index for each of the n exported goods as

well as Caves, Christensen and Diewert’s (1982) result that if the GDP function follows a translog

functional form, and the translog parameters are time invariant, then a Tornquist price index is

the exact price index of the GDP function. Using the definition of Tornquist price index, it is then

easy to compute for each good n a price index for all other goods in the economy, denoted p−n.

Equation (19) becomes

stn(ptn, p

t−n, v

t)

= (Cw − 1)a0n + (Cw − 1)ann lnptnpt−n

+M∑

m6=l,m=1

cnm

Cw−1∑c=1

ln

(vtmvtl

)c

+ µtn, ∀n. (21)

With an additive stochastic error term, µtn, to capture measurement errors, equation (21)

is the basis used for the estimation of own price effect, ann, and hence the export price elasticity

of the rest-of-the-world, e?nn.

The second problem is that we do not have enough time variation to estimate these parameters

by country, so given that we assume that the GDP functions are common up to a constant, we will

be pooling the data together and estimating the common ann using both cross-country and time

variations and introducing year and country-specific fixed effects that are all specific to each good

n. The country-specific fixed effects (for each good n) will control, for example, for the level of

trade restrictiveness in each importing country that may be correlated with the price received by

exporters, as long as trade restrictiveness does not vary significantly across time. The year fixed

effects (for each good n) will capture general shocks to world markets of good n.

15

There are also several econometric problems. Unit prices can be endogenous or measured with

error. There may also be selection bias due to the fact that some products may not be exported by

the rest-of-the-world to a particular country. Finally, there may be partial adjustment of exported

quantities to changes in prices which may lead to serial correlation in the error term.

To address all the econometric problems we follow the procedure in Kee et al. (2008). We

instrumented unit values using the simple and inverse-distance weighted averages of the unit values

of the rest-of-the-world, as well as the trade-weighted average distance of country c to all the

exporting countries of good n. We corrected for selection bias by introducing the Mills ratio of

probit equation that determines whether or not the good was exported by the rest-of-the-world

using the procedure in Semykina and Wooldridge (2010), but only when the test they propose

suggests that selection bias is a problem. We also test for serial correlation in the error term,

and, when serial correlation is present, we then estimate a dynamic model by introducing a lagged

dependent variable using the GMM system estimators developed by Arellano and Bover (1995).

This estimation strategy corresponds to Arellano and Bond (1991) difference GMM estimators,

with a level equation added to the system to improve efficiency.16

Finally, for equation (19) to be the solution of the GDP maximization problem, the second order

necessary conditions need to be satisfied (the Hessian matrix needs to be negative semi-definite).

This implies that the estimated export elasticities of the rest-of-the-world are not negative. For

this to be true for all observations:

ann ≥ sn (1− sn) (22)

where sn is the maximum share in the sample for good n. Thus, when the estimated ann does

not satisfy the curvature condition described by expression (22), we impose that the estimated

ann ≡ sn, which ensures that all elasticities are positive.

3.2 Estimating export supply elasticities from the point of view of the exporter

The export supply elasticities from the point of view of the exporter are used as instruments

for the export supply elasticity of the rest-of-the-world from the point of view of the importer.

The estimation procedure is identical to the one followed above, except for the fact that we are

16See Kee et al. 2008 for more details.

16

not summing the GDP functions of rest-of-the-world countries. In order to derive the export share

equation of each exporting country we use the first order condition of GDP maximization of each

country. The share equation to be estimated is then given by:

stn(ptn, p

t−n, v

t)

= b0n + bnn lnptnpt−n

+

M∑m 6=l,m=1

dnm

(vtmvtl

)+ utn, ∀n (23)

where b and ds are parameters to be estimated after pooling observations across countries for each

good n. The export supply elasticity of good n in each exporting country is then given by:

ex?nn ≡∂qtn

(pt, vt

)∂ptn

ptnqtn

=bnnstn

+ stn − 1 > 0 (24)

We are facing the same econometric problems and data constraints as when estimating the export

supply elasticities of the rest-of-the-world, and we therefore follow the procedure described in the

previous section.

3.3 Data

The dataset used to estimate export supply elasticities consists of export values and quanti-

ties reported by different countries to the UN Comtrade system at the six-digit level of the HS

classification (around 4600 products).17 The HS classification was introduced in 1988. The basic

data set consists of an unbalanced panel of exports for 117 countries at the six-digit level of the HS

classification for the period 1988-2009. The number of countries obviously varies across products

depending on the presence of export flows and on the availability of trade statistics using the HS

classification.

There are three factor endowments included in the regression: labor, capital stock and agricul-

tural land. Data on labor force and agricultural land are from the World Development Indicators

(WDI, 2012). Data on capital endowments is constructed using the perpetual inventory method

based on real investment data in WDI (2012).

The estimation sample did not include goods where the recorded trade value at the six-digit

level of the HS classification represented less than 0.01 percent of exports (or it had an absolute

17It is available at the World Bank through the World Integrated Trade System (WITS).

17

value of less than 50 thousand dollars). This eliminated less than 1 percent of the value of exports in

the sample, and it is necessary in order to avoid biasing our results with economically meaningless

exports. The elasticities are constructed following equation (20), where the export share is the

sample average (i.e., we constrained the elasticities to be time invariant). We also purged the

reported results from extreme values by dropping from the sample the top and bottom 1 percent

of the estimates.

3.4 Empirical Results

We have estimated a total of 268240 rest-of-the-world export supply elasticities corresponding

to 119 importers at the six-digit level of the HS classification.18 Figure 1 provides a plot of the dis-

tribution of the inverse of these rest-of-the-world supply elasticities, which captures the importer’s

market power when facing exports from the rest-of-the world. The inverse of these export supply

elasticities is also equal to the level of the optimal tariff if the importer were to use its market

power. The median of the inverse of the export supply elasticity of the rest-of-the-world is equal

to 0.044, which implies that the median optimal tariff in the world should be around 4.4 percent.

Table 2 provides the mean and standard deviation of export supply elasticities faced by each

importer in the sample used to estimate equation (15), so it excludes some countries for which

we do not have applied or bound tariffs. Moreover, we drop from Table 2 information about

individual members of the European Union given that this preferential trade agreement represents

a single decision making unity for trade policy purposes.19 The economies facing the lowest export

supply elasticities and therefore having the strongest market power are the United States and the

European Union, with average optimal tariffs above 15 percent. The countries facing the highest

export supply elasticity, and therefore being close to price-taking behavior in international markets

are Burundi, Grenada and Benin, all with average optimal tariffs below 0.001 percent.

We provide a few external tests of these estimates. First, as suggested in footnote 11, with

18We have also estimated rest-of-the-world export supply elasticities for individual members of the EuropeanUnion. If we count individual European members, we reach a total number of 317348 rest-of-the-world export supplyelasticities corresponding to 131 importers at the six-digit level of the HS classification.

19We perform the same analisys using data for individual EU members instead. The results are very similareconomically and statistically and are available upon request. In order to calculate the market power of the EU wefollowed a procedure similar to the one described in Section 3: we first estimate parameter ann using equation (21)and then, using aggregated data for EU members where we purged intra-EU trade flows, we calculate market powerwith the assistance of expression (20).

18

information on import demand elasticities and export supply elasticities for each exporter, rest-of-

the-world export supply elasticity faced by importer i can be approximated by:

e?i =1

mi/xw

ex? +∑c 6=i

emcmc

xw

(25)

where ex? is the export supply of the world, which can be approximated by the weighted sum of

export supply elasticities estimated from the exporter’s point of view, and emc is the absolute value

of the import demand elasticity of country c, which has been estimated by Kee et al. (2008). The

average and standard deviation of export supply elasticities estimated for each exporting country

are given in Table 2. The average could seem quite high, but it is important to remember that these

are export supply elasticities are estimated at the six-digit level of the HS classification keeping

all prices constant, and among these prices that are kept constant there are some which are very

close substitutes. For example, HS 010511 is the product code for live chickens under 185 grams,

and HS 010512 is for live turkeys below 185 grams. Note that in order to derive equation (25) in

footnote (11) we assumed that the export supplies were not differentiated by importer, whereas

our estimates of e? described in section 3.1 assume that export supply elasticities of the rest-of-the-

world are differentiated by destination. Thus, we do not expect our estimates to be equal to the

ones in (25).

In the first column of Table 3 we provide estimates of the correlation between our estimate

of the export supply elasticity faced by each importer, and its proxy using equation (25).20 In

the second column we split equation (25) into its three elements, specifically: the export supply

elasticity of the world for each good proxied by the weighted average export supply elasticity

of each exporter, the import-weighted import demand elasticity in the rest-of-the-world, and the

import share of the importer in world’s markets. As expected there is a positive correlation in the

first column, and Figure 2 provides a partial plot of our estimate of the export supply elasticity

faced by each importer, against the one calculated using the right-hand-side of equation (25). The

positive correlation is quite clearly illustrated in Figure 2. In the second column of Table 3, as

20Note that in order to provide estimates of the proxy using equation (25) for all six-digit HS goods, we replacedsome missing average export supply elasticities by the four-digit HS average (or the two-digit when the four-digitwas also missing). The reason is that it was impossible to estimate some export supply elasticities from the point ofview of the exporter for some products using equation (24) because there was not enough variation in the data (notenough exporters). This was not a problem when estimating the export supply elasticity faced by importers using(20) because there is always a sufficiently large number of importers.

19

expected, when decomposing (25) into its three elements, we find that both average elasticities

have a positive sign (the import demand elasticities are measured in absolute value), and the

import share a negative sign.

The second external test uses the estimates by Broda et al. (2008) of export supply elasticities

faced by importers at the six-digit level of the HS classification for thirteen countries that were not

WTO members. Thus, the third column in Table 3 provides a correlation between their estimates

and our estimates. It shows again a positive and statistically significant correlation for these thirteen

countries, which again confirms the validity of our estimates. Note again that their estimates and

ours vary in the assumptions made to obtain them, as they impose a CES structure on the demand

side, whereas our elasticities are derived from the supply side (the GDP function) and we make

no assumptions on the demand side. Thus, we shouldn’t expect them to be equal, but positively

correlated as they are both capturing export supply elasticities faced by importers.

Finally, Broda et al. (2008) provide, as an external test, a regression of the export supply

elasticities faced by the importer on the GDP of each importing country, their share in import

markets and a measure of remoteness of each importing country. Remoteness is built as the inverse

of the distance-weighted GDP of all other countries in the world. They found a negative correlation

with GDP, and the import share, suggesting that larger countries are likely to face smaller elasticities

and therefore have more market power, and a negative correlation was also found with remoteness,

suggesting that countries located far away from world markets are more likely to have more market

power, as they represent a larger share of regional demand that tends to be isolated from rest-of-

the-world demand due to larger trade costs. The fourth column of Table 3 provides the correlation

between our estimates and these three variables, and they confirm the findings by Broda et al.

(2008).

4 Evidence of non-cooperative behavior on WTO’s tariff waters

To empirically assess the degree to which WTO member countries cooperate in the presence

of tariff water we rely on the estimation of equation (15). We use data on applied MFN tariffs and

tariff bounds for the period 2000-2009.21 The econometric strategy also relies on data to control for

21This circumvents the problem that bound tariffs negotiated during the Uruguay Round were allowed a transitionperiod until 2000, which may artificially create negative or positive tariff water. The applied MFN tariffs were

20

the rest-of-the-world export supply elasticities, which are used in measuring the importer’s market

power, as well as export supply elasticity from the point of view of exporters. These estimated

elasticities were discussed in the previous Section. Table 1 provides the average and standard

deviation applied MFN and bound tariffs as well as information on tariff water across countries. It

is clear that among developed nations only Australia and New Zealand have significant amounts

of tariff water, with 7 and 9 percentage points average difference between their bound and applied

tariffs, respectively. On the other hand, most developing countries have more than 10 percentage

points of tariff water in their tariff schedules, reaching more than 40 percentage points in several

cases.

To test our two predictions we estimate equation (15) using six different specifications. In the

first specification we use our estimate of market power (1/e?). However, it is clear that these elas-

ticities are measured with errors since they are the outcome of the econometric strategy described

in Section 3. Moreover, the data described in Table 2 make it evident that our elasticities are

imprecisely estimated. In this case, the presence of outliers is evident given the size of the standard

deviation relative to the average values across countries. For these reasons, we follow Broda et al.

(2008) in considering alternative nonlinear measures of market power. The second specification

uses the log of 1/e?. The third specification uses a dummy that takes a value of 1 for goods that

are in top and middle thirds of the distribution of market power within each country. The fourth

column uses a separate dummy for the top third, and the middle third of goods in terms of market

power within each country. Broda et al. (2008) build these dummies using the definition above,

but one could argue that the top third goods in terms of market power in Burundi may well be at

the bottom of the market power distribution when considering all countries and goods. Thus, the

fifth specification uses a dummy that takes a value of 1 when the market power of that country in

that particular good is on the top or middle thirds of the world distribution of market power. The

last specification splits the dummy into two dummies that capture the top third and the middle

third separately, as in the fourth specification.

Table 4 presents the OLS results of (15), which broadly confirm our first prediction: importer’s

market power is negatively correlated with applied tariffs in the absence of water. With the ex-

ception of the specification in the first column, all coefficients on importer’s market power are

obtained using the World Integrated Trade System (WITS) while tariff bounds negotiated during the Uruguay roundwere provided by the WTO.

21

statistically significant. Results also confirm the second prediction: the interaction term between

water and importer’s market power is positive and statistically significant, with the exception of

the specification in the first column. This suggests that in the presence of sufficiently deep tariff

water, countries tend to use their market power.

The degree of tariff water needed for this to be the case varies across specifications between

19 and 67 percentage points. These could seem like extreme cases but as can be seen from Table

1 there are a significant number of countries which have an average level of tariff water above 19

percent. Our dataset suggests that about 33 percent of the observations in the sample used in Table

4 have tariff water levels above 19 percentage points, while about 2.6 percent have tariff water levels

above 63 percentage points. Likewise, the results displayed in the fourth and sixth columns strongly

suggest that the degree of water needed for countries to use their market power is lower for goods in

which countries have high market power than for goods in which countries display medium or low

market power. More importantly, the statistical confirmation of our predictions indicates that the

WTO does induce cooperation among member countries to a large extent, but also suggests that

the presence of deep levels of tariff water may contribute to deviations from cooperative behavior.

The estimates in table 4 could suffer from endogeneity bias as discussed in section 2. Thus, table

5 presents results instrumenting for tariff water and market power. Water is instrumented using

measures of water vapour, and market power is instrumented using the exogenous right-hand-side

variables in expression (25): the average import demand elasticity in the rest-of-the-world, and

the world’s export supply elasticity (although the latter is perfectly collinear with the HS six-digit

fixed effects and therefore drops from the list of instruments).22 The interaction of water with the

importer’s market power is instrumented using the interaction of water vapour with the average

import demand elasticity in the rest-of-the-world, and the interaction of water vapour with the

world’s export supply elasticity (ex?). Note that the number of instruments is larger than the

number of endogenous variables, which will allows us to test for the validity of the instruments

using an over-identification test.

The results in table 5 confirm again the first prediction: importer’s market power is negatively

correlated with applied tariffs in the absence of tariff water. The coefficient on the importer’s

market power is statistically significant across specifications, except in the third column. Results

22We do not use the import share of the importer in world’s trade which appears on the right-hand-side of equation(25) because this is likely to be endogenous to applied tariffs.

22

also tend to confirm our second prediction: importers use their market power in setting trade policy

in the presence of large amounts of tariff water. However, the coefficient on the interaction between

tariff water and market power is not statistically significant in specifications 1 and 2. Note that

in columns 1 and 2 we cannot reject the null that we are in the presence of weak instrument,

which may bias our results and explain the statistically insignificant coefficients. The results shown

in column 5 and 6 (our preferred specification) suggest that the degree of tariff water needed for

importers to use their market power in setting tariffs has to be above 26 percentage points. This

implies that significant levels of tariff water are required for the presence of non-cooperative tariff

policy behavior under the WTO.23

4.1 Fear of Retaliation

The results above suggest that WTO members tend to behave more cooperatively than is

legally required. Why is it that they do not use their market power when there are no legal

constraints? A potential explanation is fear of retaliation by trading partners. Consider a country

with a significant amount of tariff water in its tariff schedule. When evaluating whether or not to

raise its tariffs to non-cooperative levels, the cost of retaliatory trade measures by trading partners

with significant amounts of market power and tariff water would have to be weighted against the

terms-of-trade gains associated with the non-cooperative tariff.

Blonigen and Bown (2003) and Bown (2008) have shown that retaliation threats will make

importers less likely to impose antidumping measures, and more likely to behave cooperatively

within its WTO’s legal commitments. In order to explore whether fear of retaliation can make

WTO members behave more cooperatively outside WTO’s legal commitments, i.e., within WTO

tariff waters, we first need to construct a measure of fear of retaliation, and then check whether

importers are more prone to use their market power within their tariff waters when they have little

fear of retaliation by their trading partners.

Let’s denote the fear of retaliation in country c by Fc which, by construction, does not vary

across tariff lines, as trading partners do not necessarily retaliate within the same tariff line, but can

23These results are broadly confirmed when using an IV between estimator instead of the within estimator usedfor the results reported in Table 5. Indeed our main source of variation is across HS six-digit lines and within HStwo-digit for each country and year, and therefore the between estimator provides very similar results. Results ofTables 4 and 5 are also confirmed when using data for individual European countries. The results are similar to theresults described in Tables 4-5 and are available upon request.

23

retaliate across their entire import bundle.24 We define fear of retaliation as the average maximum

increase in tariffs in partner countries that would lead to the same decline in country c’s value

of exports than if all partners were to increase their current applied tariffs to their bound levels.

This definition is similar in spirit to the one used to define trade restrictiveness in Anderson and

Neary (1996, 2003). To apply their concept we use the partial equilibrium approach developed by

Feenstra (1995) and used by Kee et al. (2009) to measure trade restrictiveness. We denote country

c’s partner countries by subscript j while we continue using subscript p to identify products. Fear

of retaliation in country c is then defined as:

Fc : ∆

∑p

∑j

p?p,jmp,j,c

∆tp,j=Wp,j

= ∆

∑p

∑j

p?p,jmp,j,c

∆tp,j=Fc

(26)

where mp,j,c represents country j’s imports of product p from country c. Notice that on the

right-hand side of (26) we index the world price by product p, given that we allow for products of

type p imported by different countries to be heterogenous, and assume that all countries change

their applied MFN tariffs to the same uniform tariff that replicates the change in imports from

country c which is described in the left-hand-side of this expression.

Totally differentiating both sides of the equality in (26), noting that by definition the change in

partner tariffs on the left hand side is equal to the extent of water available in their tariff schedule,

allows us to solve for the fear of retaliation in country c, Fc. Before solving for Fc, note that the

marginal change in world prices faced by importer j following a change on its MFN tariff on each

good p (assuming goods from different sources are homogenous) is given by:25

∂p?p,j∂tp,j

=εmp,j

(1 + tp,j)(ε?p,j + εmp,j

) (27)

where εmp,j > 0 is the absolute value of the import demand elasticity of good p in partner j.

Differentiating (26) with respect to changes in partner tariffs tp,j , using (27), and solving for Fc

yields (while taking the absolute value of the changes in exports):

24There are some well known anecdotal examples of this. In 1999, the United States impose 100 percent tariffs onnine different goods imported from Europe going from pecorino cheese to cashmere clothing, as retaliation for theEU’s banana regime.

25This is obtained by starting from the identity between total imports of good p by country j being equal to totalexports of the rest of the world of good p to country country j, and then differentiating.

24

Fc =

∑p

∑j mp,j,c

εmp,j(1+tp,j)(ε?p,j+εmp,j)

(1 + εx?p,c

)Wp,j∑

p

∑j mp,j,c

εmp,j(1+tp,j)(ε?p,j+εmp,j)

(1 + εx?p,c

) . (28)

where εx?p,c is the export supply elasticity of country c as an exporter of product p. The comparative

statistics are quite clear. If the importing partner country has a lot of market power (i.e., a small

ε?p,j), then the tariff water (Wp,j) on exports of that good from that partner has a greater weight

in our measure of fear of retaliation. Similarly, the strongest is the import demand of the partner,

the larger the weight given to water in that partner’s product. The same is true for exports to that

partner as well as the export supply elasticity of country c of product p.

Expression (28) enables us to calculate the fear of retaliation faced by each country in the

sample used in Table 5 and this allows us to split the sample into three equally numbered group of

countries with low, medium and high fear of retaliation. We use data for the year of 2006 to estimate

(28) since this is the year prior to the financial crisis. We then proceed by estimating expression

(15) separately for countries with low and high fear of retaliation. We expect that the variable

measuring the product between market power and water is positive and statistically significant for

the group of countries with low fear of retaliation, while we expect that the interaction between

power and water is not statistically significant in the case of countries with high fear of retaliation.

Table 6 shows the IV estimates of expression (15) for the sample of countries with low and high

fear of retaliation separately. We use two measures of market power: the log of the inverse export

supply elasticity in columns 1 and 2, and a dummy that takes a value of 1 for observations that

are in the top and middle third of the distribution of market power across all countries and goods

in columns 3 and 4. The first prediction which suggests that in the absence of tariff water market

power is negatively correlated with applied tariffs is confirmed for countries in both the low and

high fear of retaliation samples, regardless of the measure of market power we used, except perhaps

in column 3 where the negative coefficient is not statistically significant.

More interestingly, the use of market power in the presence of large amounts of tariff waters

is only present for countries with low fear of retaliation. Indeed, the coefficient on the interaction

of market power and tariff water is positive and significant only in the sample of countries which

have low fear of retaliation. In the case of countries which have more to lose from retaliation by

their trading partners, there is no evidence of non-cooperative tariff setting as the coefficient on the

25

interaction between market power and tariff water is either negative or not statistically significant.

5 Concluding Remarks

In this paper, we investigate whether WTO members set their applied tariffs non-cooperatively

when they have the discretion to do so. More precisely, we examine the extent to which WTO

members use their market power when their tariff bounds are significantly above their applied tariffs,

i.e., in the presence of tariff water. We develop a simple model of tariff setting that suggests that

in the absence of tariff water, cooperative tariffs will be negatively correlated with the importer’s

market power. Indeed, the more market power the importer has, the stronger are the incentives

to negotiate lower tariffs. On the other hand, in the presence of tariff-water, non-cooperative tariff

setting is possible and the model predicts a positive correlation between the importer’s applied

tariffs and market power in international markets.

To test these predictions, we first estimate the degree of market power (the inverse of rest-of-

the-world export supply elasticity faced by the importer) at the tariff line level for more than 100

WTO member countries. Our econometric approach is based on Kholi’s (1991) revenue function

approach, and is sufficiently flexible to allow us to also estimate export supply elasticity for each

exporter.

We use then our elasticity estimates to study the effects of market power on trade policy with

and without the presence of tariff water. Because tariff water and market power may be endogenous

we use an instrumental variable approach, where the extent of tariff water above the prohibitive

tariff (water vapour), the average import demand elasticity in the rest-of-the-world, the export

supply elasticity of the world are used as instruments.

Results are in line with our theoretical predictions. First, we find that importer’s market power

tends to be negatively correlated with applied tariffs in the absence of tariff water. Second, our

results suggest that in the presence of deep tariff water (i.e., levels above 20 percentage points),

importing countries employ their market power in setting applied tariffs. These results are robust

to different measures of market power, econometric models and instrumental variables.

An important corollary result is that in the presence of moderate levels of tariff water, WTO

members tend to set their tariffs cooperatively. This suggests that cooperation is present within

26

WTO tariff waters. Thus, in spite of significant amounts of policy space provided by tariff water,

WTO members tend to behave cooperatively when setting their trade policy.

One explanation for cooperative behavior in the absence of legal constraints is the fear of

retaliation by trading partners with significant amounts of market power and tariff water. We show

that WTO members which have little to lose from retaliation tend to set tariffs non-cooperatively

within their tariff waters, while WTO members that may have more to lose in case of retaliation

are more likely to set tariffs cooperatively within their tariff waters.

27

References

[1] Amador, Manuel, and Kyle Bagwell. 2012. “Tariff Revenue and Tariff Caps”, American Eco-

nomic Review 102(3): 459–65.

[2] Anderson, James and Peter Neary, 1996. “A New Approach to Evaluating Trade Policy”,

Review of Economics and Statistics 63, 107-125.

[3] Anderson, James and Peter Neary, 2003. “The Mercantilist Index of Trade Policy”, Interna-

tional Economic Review 44, 627-649.

[4] Arellano, Manuel and Stephen Bond, 1991. “Some Tests of Specification for Panel Data: Monte

Carlo Evidence and an Application to Employment Equations”, Review of Economic Studies

58, 277-297.

[5] Arellano, Manuel and Olympia Bover, 1995. “Another look at the instrumental variable esti-

mation of error-components models”, Journal of Econometrics 68, 29-51.

[6] Bagwell, Kyle and Robert W. Staiger, 1999. “An Economic Theory of GATT”, American

Economic Review 89(1): 215-248.

[7] Bagwell, Kyle and Robert W. Staiger, 2011. “What Do Trade Negotiators Negotiate About?

Empirical Evidence from the World Trade Organization”, American Economic Review 101(4):

1238-73.

[8] Beshkar, Mostafa, Eric Bond, and Youngwoo Rho, 2012. “Tariff Binding and Overhang: The-

ory and Evidence”, working paper.

[9] Blonigen, Bruce and Chad Bown, 2003. “Antidumping and retaliation threats”, Journal of

International Economics 60: 249-273.

[10] Bown, Bown, 2003. “On the economic success of GATT/WTO dispute settlement”, Review of

Economics and Statistics 86: 811-823.

[11] Broda, Christian, Nuno Limao, and David E. Weinstein, 2008. “Optimal Tariffs and Market

Power: The Evidence”, American Economic Review 98(5): 2032-65.

28

[12] Caves, Douglas W., Laurits R. Christensen, and Erwin Diewert, 1982. “Multilateral Com-

parisons of Output, Input, and Productivity using Superlative Index Numbers,” Economic

Journal 92, 73-86.

[13] Croome, John, 1995. Reshaping the World Trading System: a History of the Uruguay Round.

World Trade Organization.

[14] Eicher, Theo S. and Christian Henn, 2011. “In Search of WTO Trade Effects: Preferential

Trade Agreements Promote Trade Strongly, but Unevenly” Journal of International Economics

83, 137-153.

[15] Feenstra, Robert, 1995. “Estimating the Effects of Trade Policy”, Handbook of International

Economics vol. 3.

[16] Foletti, Liliana, Marco Fugazza, Alessandro Nicita and Marcelo Olarreaga, 2011. “Smoke in

the (Tariff) Water”, The World Economy 34(2): 248-264.

[17] Grossman, Gene M., and Elhanan Helpman, 1995. “Trade wars and trade talks”, Journal of

Political Economy 103(4): 675-708.

[18] Harrigan, James, 1997. “Technology, Factor Supplies, and International Specialization: Esti-

mating the Neoclassical Model,” The American Economic Review 87 (4), 475-494.

[19] Hoekman, Bernard and Michael Kostechi, 2009. The Political Economy of the World Trading

System, Oxford University Press, Third Edition.

[20] Horn, Henrik, Giovanni Maggi and Robert Staiger, 2010. “Trade Agreements as Endogenously

Incomplete Contracts”, The American Economic Review 100 (1), 394-419.

[21] Kee, Hiau Looi, Alessandro Nicita and Marcelo Olarreaga, 2008. “Import Demand Elasticities

and Trade Distortions”, The Review of Economics and Statistics 90(4): 666-682.

[22] Kohli, Ulrich, 1991. Technology, Duality, and Foreign Trade: The GNP Function Approach to

Modeling Imports and Exports, The University of Michigan Press, Ann Arbor.

[23] Gawande, Kishore, Pravin Krishna and Marcelo Olarreaga, 2012. “Lobbying competition over

trade policy”, International Economic Review 53 (1), 115-132.

29

[24] Olarreaga, Marcelo, Isidro Soloaga, and Alan Winters, 1999. “What’s behind MERCOSUR’s

common external tariff?”, Policy Research Working Paper Series 2231, The World Bank.

[25] Rauch, James, 1999. “Network versus markets in international trade”, Journal of International

Economics 48, 7-35.

[26] Rose, Andrew K., 2004. “Do we really Know that the WTO Increases Trade?” American

Economic Review 13 (4): 682-698.

[27] Semykina, Anastasia and Jeffrey Wooldridge, 2010. “Estimating Panel Data Models in the

Presence of Endogeneity and Selection: Theory and Application,” Journal of Econometrics

157(2), 375-380.

[28] Tomz, Michael, Judith L. Goldstein, and Douglas Rivers, 2007. “Do we really Know that the

WTO Increases Trade? Comment.” American Economic Review 97 (5), 2005-2018.

30

Table 1: Descriptive statistics of tariffs and water

Country Stat Tariff MFN Tariff Vapourbound applied water water

Antigua and Barbuda Mean 0.720 0.149 0.571 0.013Std. Dev. 0.323 0.093 0.301 0.057

Argentina Mean 0.319 0.127 0.193 0.014Std. Dev. 0.063 0.071 0.083 0.048

Australia Mean 0.110 0.042 0.068 0.003Std. Dev. 0.116 0.053 0.077 0.020

Bahrain Mean 0.344 0.067 0.280 0.004Std. Dev. 0.143 0.109 0.076 0.026

Bangladesh Mean 1.465 0.167 1.300 0.499Std. Dev. 0.776 0.118 0.744 0.578

Barbados Mean 0.810 0.158 0.654 0.030Std. Dev. 0.273 0.219 0.262 0.159

Belize Mean 0.603 0.127 0.476 0.004Std. Dev. 0.200 0.119 0.198 0.032

Benin Mean 0.229 0.135 0.128 0.004Std. Dev. 0.239 0.067 0.200 0.033

Bolivia Mean 0.399 0.087 0.313 0.005Std. Dev. 0.009 0.034 0.034 0.031

Botswana Mean 0.224 0.103 0.121 0.008Std. Dev. 0.240 0.122 0.224 0.130

Brazil Mean 0.312 0.139 0.174 0.018Std. Dev. 0.076 0.065 0.081 0.053

Brunei Mean 0.254 0.028 0.226 0.002Std. Dev. 0.084 0.058 0.071 0.015

Bulgaria Mean 0.254 0.079 0.175 0.003Std. Dev. 0.160 0.082 0.137 0.028

Burkina Faso Mean 0.306 0.119 0.216 0.008Std. Dev. 0.391 0.066 0.359 0.055

Burundi Mean 0.555 0.218 0.404 0.022Std. Dev. 0.444 0.130 0.392 0.110

Cote d’Ivoire Mean 0.097 0.121 0.015 0.001Std. Dev. 0.068 0.068 0.045 0.021

Cameroon Mean 0.800 0.216 0.584 0.050Std. Dev. 0.000 0.099 0.099 0.124

Canada Mean 0.052 0.040 0.013 0.000Std. Dev. 0.053 0.054 0.022 0.002

Central African Rep. Mean 0.372 0.168 0.204 0.000Std. Dev. 0.103 0.088 0.119 0.000

Chile Mean 0.252 0.066 0.186 0.005Std. Dev. 0.029 0.010 0.031 0.027

China Mean 0.099 0.116 0.002 0.000Std. Dev. 0.073 0.092 0.014 0.002

Colombia Mean 0.414 0.126 0.288 0.022Std. Dev. 0.209 0.068 0.202 0.112

Congo Mean 0.231 0.191 0.040 0.000Std. Dev. 0.102 0.103 0.078 0.000

Costa Rica Mean 0.425 0.059 0.366 0.009Std. Dev. 0.120 0.079 0.119 0.048

Croatia Mean 0.064 0.068 0.011 0.000Std. Dev. 0.054 0.066 0.024 0.001

Czeck Republic Mean 0.048 0.047 0.002 0.000Std. Dev. 0.062 0.062 0.011 0.000

Dominica Mean 0.705 0.141 0.565 0.012Std. Dev. 0.327 0.168 0.275 0.078

Egypt Mean 0.296 0.138 0.168 0.012Std. Dev. 0.673 0.612 0.307 0.249

El Salvador Mean 0.359 0.070 0.289 0.007Std. Dev. 0.128 0.085 0.123 0.040

Estonia Mean 0.091 0.030 0.063 0.000Std. Dev. 0.077 0.062 0.062 0.007

European Union Mean 0.044 0.044 0.001 0.000Std. Dev. 0.044 0.044 0.008 0.002

Gabon Mean 0.224 0.182 0.084 0.001Std. Dev. 0.167 0.095 0.128 0.012

Georgia Mean 0.064 0.053 0.020 0.000Std. Dev. 0.058 0.056 0.039 0.002

Ghana Mean 0.845 0.164 0.681 0.053Std. Dev. 0.264 0.094 0.244 0.160

Grenada Mean 0.599 0.138 0.461 0.006Std. Dev. 0.229 0.093 0.231 0.047

Guatemala Mean 0.415 0.063 0.352 0.015Std. Dev. 0.171 0.069 0.169 0.076

Guinea Mean 0.164 0.129 0.067 0.000Std. Dev. 0.143 0.069 0.108 0.004

Guyana Mean 0.555 0.096 0.460 0.001Std. Dev. 0.157 0.083 0.159 0.007

Honduras Mean 0.309 0.067 0.242 0.002Std. Dev. 0.088 0.071 0.096 0.018

Country Stat Tariff MFN Tariff Uselessbound applied water water

Hungary Mean 0.067 0.063 0.005 0.000Std. Dev. 0.082 0.077 0.023 0.004

Iceland Mean 0.168 0.040 0.128 0.007Std. Dev. 0.205 0.063 0.187 0.064

India Mean 0.441 0.222 0.225 0.039Std. Dev. 0.353 0.172 0.292 0.174

Indonesia Mean 0.372 0.067 0.306 0.024Std. Dev. 0.123 0.086 0.123 0.081

Israel Mean 0.204 0.042 0.162 0.033Std. Dev. 0.400 0.104 0.381 0.225

Jamaica Mean 0.525 0.087 0.439 0.014Std. Dev. 0.224 0.111 0.205 0.067

Japan Mean 0.031 0.032 0.001 0.000Std. Dev. 0.048 0.047 0.009 0.000

Jordan Mean 0.169 0.151 0.040 0.000Std. Dev. 0.152 0.154 0.073 0.007

Kenya Mean 0.941 0.209 0.733 0.064Std. Dev. 0.188 0.166 0.207 0.172

Korea Mean 0.153 0.109 0.048 0.003Std. Dev. 0.356 0.336 0.081 0.045

Kyrgyzstan Mean 0.064 0.038 0.029 0.000Std. Dev. 0.047 0.049 0.037 0.000

Latvia Mean 0.078 0.038 0.041 0.000Std. Dev. 0.095 0.056 0.082 0.010

Lesotho Mean 0.996 0.118 0.878 0.208Std. Dev. 0.631 0.119 0.642 0.385

Lithuania Mean 0.066 0.038 0.031 0.000Std. Dev. 0.067 0.060 0.051 0.007

Madagascar Mean 0.246 0.105 0.144 0.001Std. Dev. 0.066 0.068 0.078 0.008

Malawi Mean 0.772 0.105 0.666 0.036Std. Dev. 0.397 0.099 0.356 0.103

Malaysia Mean 0.150 0.086 0.067 0.002Std. Dev. 0.123 0.102 0.098 0.034

Mali Mean 0.201 0.120 0.112 0.001Std. Dev. 0.214 0.065 0.185 0.016

Malta Mean 0.493 0.057 0.435 0.005Std. Dev. 0.095 0.041 0.100 0.036

Mauritius Mean 0.865 0.099 0.776 0.110Std. Dev. 0.491 0.166 0.465 0.197

Mexico Mean 0.351 0.152 0.200 0.010Std. Dev. 0.046 0.094 0.090 0.041

Mongolia Mean 0.184 0.044 0.141 0.000Std. Dev. 0.050 0.018 0.052 0.003

Morocco Mean 0.403 0.248 0.178 0.007Std. Dev. 0.139 0.204 0.173 0.052

Namibia Mean 0.255 0.111 0.144 0.012Std. Dev. 0.293 0.129 0.283 0.179

New Zealand Mean 0.117 0.034 0.083 0.002Std. Dev. 0.116 0.044 0.080 0.014

Nicaragua Mean 0.423 0.058 0.365 0.002Std. Dev. 0.099 0.074 0.096 0.026

Niger Mean 0.428 0.130 0.316 0.023Std. Dev. 0.437 0.069 0.413 0.126

Nigeria Mean 0.949 0.152 0.797 0.168Std. Dev. 0.516 0.210 0.459 0.311

Norway Mean 0.033 0.013 0.022 0.000Std. Dev. 0.041 0.042 0.031 0.005

Oman Mean 0.135 0.061 0.077 0.003Std. Dev. 0.172 0.085 0.116 0.046

Panama Mean 0.232 0.081 0.153 0.002Std. Dev. 0.115 0.085 0.101 0.018

Papua New Guinea Mean 0.333 0.040 0.293 0.003Std. Dev. 0.145 0.094 0.132 0.027

Paraguay Mean 0.326 0.117 0.210 0.003Std. Dev. 0.067 0.068 0.086 0.022

Peru Mean 0.302 0.096 0.206 0.007Std. Dev. 0.026 0.058 0.061 0.031

Philippines Mean 0.248 0.055 0.194 0.009Std. Dev. 0.114 0.061 0.099 0.043

Poland Mean 0.075 0.075 0.001 0.000Std. Dev. 0.112 0.113 0.008 0.000

Romania Mean 0.044 0.084 0.002 0.000Std. Dev. 0.046 0.090 0.009 0.000

Rwanda Mean 0.873 0.177 0.709 0.044Std. Dev. 0.283 0.111 0.280 0.132

Saint Kitss Mean 0.818 0.141 0.677 0.011Std. Dev. 0.243 0.103 0.230 0.057

Saint Lucia Mean 0.746 0.136 0.610 0.024Std. Dev. 0.350 0.121 0.328 0.082

32

Country Stat Tariff MFN Tariff Uselessbound applied water water

Saudi Arabia Mean 0.107 0.063 0.051 0.001Std. Dev. 0.062 0.040 0.047 0.009

Senegal Mean 0.299 0.125 0.174 0.001Std. Dev. 0.009 0.068 0.068 0.011

Singapore Mean 0.070 0.000 0.070 0.001Std. Dev. 0.040 0.000 0.040 0.007

Slovakia Mean 0.055 0.120 0.014 0.000Std. Dev. 0.070 0.151 0.048 0.014

Slovenia Mean 0.123 0.073 0.058 0.001Std. Dev. 0.112 0.063 0.082 0.013

South Africa Mean 0.195 0.085 0.110 0.012Std. Dev. 0.234 0.116 0.216 0.132

Sri Lanka Mean 0.224 0.087 0.142 0.003Std. Dev. 0.193 0.133 0.134 0.029

Swaziland Mean 0.242 0.115 0.127 0.004Std. Dev. 0.205 0.125 0.184 0.057

Tanzania Mean 1.200 0.233 0.967 0.140Std. Dev. 0.000 0.160 0.160 0.254

Thailand Mean 0.255 0.131 0.139 0.006Std. Dev. 0.139 0.145 0.119 0.042

Togo Mean 0.800 0.169 0.631 0.006Std. Dev. 0.000 0.053 0.053 0.051

Trinidad and Tobago Mean 0.577 0.085 0.492 0.015Std. Dev. 0.193 0.104 0.172 0.072

Tunisia Mean 0.495 0.255 0.241 0.009Std. Dev. 0.317 0.246 0.235 0.075

Turkey Mean 0.239 0.080 0.160 0.009Std. Dev. 0.270 0.177 0.190 0.071

Uganda Mean 0.698 0.140 0.559 0.044Std. Dev. 0.158 0.145 0.154 0.129

United Arab Emirates Mean 0.158 0.049 0.109 0.015Std. Dev. 0.240 0.057 0.213 0.139

United States Mean 0.040 0.042 0.000 0.000Std. Dev. 0.122 0.122 0.003 0.000

Uruguay Mean 0.315 0.128 0.188 0.004Std. Dev. 0.065 0.068 0.086 0.027

Venezuela Mean 0.358 0.134 0.223 0.012Std. Dev. 0.133 0.070 0.136 0.057

Zambia Mean 0.886 0.130 0.756 0.065Std. Dev. 0.411 0.109 0.353 0.169

Zimbabwe Mean 0.633 0.186 0.485 0.106Std. Dev. 0.680 0.186 0.596 0.264

Source: World Bank’s WITS at wits.worldbank.org, and Foletti et al. (2011) for the definition of water vapour.

33

Table 2: Descriptive statistics of trade elasticities

Country Statistic Import Export ROW exportdemand supply supply elas-elasticity elasticity ticity (e?)

Antigua and Barbuda Mean 1.65 24.3 694Std. Dev. 2.13 54.7 1931

Argentina Mean 1.52 27.0 99Std. Dev. 1.97 103.4 1082

Australia Mean 1.64 27.0 35Std. Dev. 2.26 101.9 141

Bahrain Mean 1.53 23.7 324Std. Dev. 1.84 63.3 733

Bangladesh Mean 1.55 52.6 157Std. Dev. 2.08 151.6 468

Barbados Mean 1.51 22.9 692Std. Dev. 1.83 56.8 2260

Belize Mean 1.63 22.5 775Std. Dev. 1.96 56.6 2376

Benin Mean 1.71 29.5 1135Std. Dev. 2.19 56.9 3942

Bolivia Mean 1.54 22.8 463Std. Dev. 2.00 87.4 1619

Botswana Mean 1.61 25.5 462Std. Dev. 2.08 93.6 1775

Brazil Mean 1.58 26.7 51Std. Dev. 2.14 100.2 144

Brunei Mean 1.59 26.2 363Std. Dev. 2.14 82.2 1853

Bulgaria Mean 1.54 22.3 155Std. Dev. 1.97 72.0 515

Burkina Faso Mean 1.81 22.4 683Std. Dev. 2.27 53.2 1743

Burundi Mean 1.89 39.0 1569Std. Dev. 3.14 84.6 5562

Cote d’Ivoire Mean 1.54 27.9 494Std. Dev. 2.07 79.0 1494

Cameroon Mean 1.73 42.6 224Std. Dev. 2.11 90.1 371

Canada Mean 1.68 27.0 16Std. Dev. 2.38 108.5 63

Central African Rep. Mean 1.52 30.8 906Std. Dev. 2.00 55.1 3325

Chile Mean 1.61 25.6 124Std. Dev. 2.15 92.0 1097

China Mean 1.62 26.4 29Std. Dev. 2.27 99.9 106

Colombia Mean 1.58 23.4 136Std. Dev. 2.12 84.1 784

Congo Mean 0.84 70.7 367Std. Dev. 0.29 109.4 591

Costa Rica Mean 1.53 26.9 239Std. Dev. 1.98 102.1 791

Croatia Mean 1.64 26.9 176Std. Dev. 2.20 100.3 1856

Cyprus Mean 1.52 27.6 251Std. Dev. 1.84 105.7 722

Czeck Republic Mean 1.64 25.7 53Std. Dev. 2.26 98.0 206

Dominica Mean 1.59 26.9 1053Std. Dev. 1.70 63.0 4084

Egypt Mean 1.54 23.1 128Std. Dev. 2.00 89.8 341

El Salvador Mean 1.56 25.5 274Std. Dev. 2.11 89.9 901

Estonia Mean 1.64 26.7 238Std. Dev. 2.30 96.3 1661

European Union Mean 5.51 47 4.64Std. Dev. 7.84 228 10.89

Gabon Mean 1.54 23.6 498Std. Dev. 1.93 87.0 1463

Georgia Mean 1.74 26.4 420Std. Dev. 2.51 70.8 1460

Ghana Mean 1.76 45.4 153Std. Dev. 2.47 90.3 347

Grenada Mean 1.80 24.6 1346Std. Dev. 2.28 49.7 5870

Guatemala Mean 1.58 25.9 240Std. Dev. 2.13 101.3 765

Guinea Mean 1.59 29.3 869Std. Dev. 1.90 69.1 2786

Guyana Mean 1.55 20.7 579Std. Dev. 1.71 52.2 1771

Country Statistic Import Export ROW exportdemand supply supply elas-elasticity elasticity ticity (e?)

Honduras Mean 1.63 27.3 383Std. Dev. 2.24 97.8 1477

Hungary Mean 1.61 26.0 63Std. Dev. 2.19 94.7 243

Iceland Mean 1.56 31.0 345Std. Dev. 2.21 116.7 1736

India Mean 1.61 24.8 50Std. Dev. 2.22 101.8 165

Indonesia Mean 1.65 26.4 77Std. Dev. 2.27 106.7 289

Israel Mean 1.59 26.5 90Std. Dev. 2.33 100.2 611

Jamaica Mean 1.61 23.7 346Std. Dev. 2.10 83.9 1189

Japan Mean 1.59 24.5 13Std. Dev. 2.27 90.0 37

Jordan Mean 1.60 24.5 265Std. Dev. 2.06 87.2 716

Kenya Mean 1.68 59.3 226Std. Dev. 2.09 142.9 541

Korea Mean 1.63 25.7 28Std. Dev. 2.28 95.1 94

Kyrgyzstan Mean 1.64 26.8 461Std. Dev. 1.88 80.4 2022

Latvia Mean 1.62 25.6 229Std. Dev. 2.24 83.2 740

Lesotho Mean 1.81 25.2 305Std. Dev. 2.14 53.7 793

Lithuania Mean 1.62 28.0 180Std. Dev. 2.18 107.4 638

Madagascar Mean 1.58 27.0 574Std. Dev. 2.13 71.8 1384

Malawi Mean 1.82 40.8 415Std. Dev. 2.65 96.2 1180

Malaysia Mean 1.69 24.5 63Std. Dev. 2.42 88.0 243

Mali Mean 1.73 22.9 532Std. Dev. 2.01 50.8 1272

Malta Mean 1.54 25.0 415Std. Dev. 2.01 95.4 1194

Mauritius Mean 1.60 32.9 337Std. Dev. 2.11 67.6 1427

Mexico Mean 1.64 26.3 29Std. Dev. 2.32 99.7 92

Mongolia Mean 1.68 23.7 531Std. Dev. 2.07 105.9 1560

Morocco Mean 1.61 25.4 152Std. Dev. 2.19 93.9 428

Namibia Mean 1.55 27.9 381Std. Dev. 2.02 97.6 973

New Zealand Mean 1.61 28.1 100Std. Dev. 2.16 102.4 574

Nicaragua Mean 1.52 24.1 475Std. Dev. 1.89 87.7 1131

Niger Mean 1.65 23.7 830Std. Dev. 2.03 47.3 2713

Nigeria Mean 1.97 39.4 66Std. Dev. 3.01 125.0 123

Oman Mean 1.63 23.0 236Std. Dev. 2.15 69.1 758

Panama Mean 1.51 24.4 301Std. Dev. 1.84 81.5 1440

Papua New Guinea Mean 1.59 21.8 480Std. Dev. 2.02 59.2 1373

Paraguay Mean 1.52 21.2 352Std. Dev. 2.01 56.5 1239

Peru Mean 1.53 24.7 214Std. Dev. 2.02 91.4 1977

Philippines Mean 1.68 27.7 115Std. Dev. 2.39 101.6 621

Poland Mean 1.62 25.8 39Std. Dev. 2.21 102.0 183

Romania Mean 1.62 26.6 79Std. Dev. 2.20 105.3 235

Rwanda Mean 1.66 26.6 826Std. Dev. 2.06 74.5 2774

Saint Kitss Mean 1.68 23.1 858Std. Dev. 1.96 55.2 2683

35

Country Statistic Import Export ROW exportdemand supply supply elas-elasticity elasticity ticity (e?)

Saint Lucia Mean 1.52 24.7 771Std. Dev. 1.64 53.4 2168

Saudi Arabia Mean 1.69 26.1 64Std. Dev. 2.40 102.4 176

Senegal Mean 1.65 20.8 457Std. Dev. 2.00 53.6 1274

Singapore Mean 1.62 31.0 42Std. Dev. 2.38 116.3 280

Slovakia Mean 1.59 25.3 93Std. Dev. 2.15 87.0 337

Slovenia Mean 1.60 27.8 151Std. Dev. 2.15 102.1 1361

South Africa Mean 1.60 24.1 67Std. Dev. 2.13 88.3 273

Sri Lanka Mean 1.72 34.1 208Std. Dev. 2.21 92.5 784

Swaziland Mean 1.60 23.8 608Std. Dev. 1.89 65.5 2031

Tanzania Mean 2.00 51.1 182Std. Dev. 3.10 99.2 327

Thailand Mean 1.59 30.5 79Std. Dev. 2.10 120.2 448

Togo Mean 1.53 41.7 341Std. Dev. 1.98 67.3 706

Trinidad and Tobago Mean 1.56 24.2 415Std. Dev. 2.01 96.3 1893

Tunisia Mean 1.65 26.8 131Std. Dev. 2.30 78.3 356

Uganda Mean 1.97 41.5 330Std. Dev. 2.86 79.3 789

United Arab Emirates Mean 1.72 22.8 55Std. Dev. 2.57 78.9 134

United States Mean 1.41 28.7 3Std. Dev. 1.95 155.2 11

Uruguay Mean 1.49 26.7 290Std. Dev. 1.81 91.2 659

Venezuela Mean 1.59 23.9 125Std. Dev. 2.19 86.4 860

Zambia Mean 1.95 39.4 236Std. Dev. 2.65 71.9 631

Zimbabwe Mean 1.48 36.2 365Std. Dev. 1.71 82.8 953

Source: Authors’ estimation and Kee et al. (2008) for import demand elasticities.

36

Table 3: External tests of the estimates of export supply elasticities faced by importers

(1) (2) (3) (4)

Log of Export supply elasticity of ROW 0.222∗∗

(left-hand-side of equation (25)) ( 0.002 )

Log of world’s export supply elasticity 0.024∗∗

(right-hand-side of equation (25)) (0.003)

Log of import demand elasticity of ROW 0.090∗∗

(right-hand-side of equation (25)) (0.004)

Log of import share -0.370∗∗ -0.421∗∗

(right-hand-side of equation (25)) (0.002) (0.003)

Log of Export supply elasticity of ROW 0.029∗∗

(Broda et al. (2008) estimates) (0.006)

Log of GDP -0.050∗∗

(0.002)

Log of remoteness -0.179∗∗

(inverse of distance-weigthed GDP of ROW) (0.012)

R2-adjusted 0.139 0.164 0.249 0.505Number of observations 268240 268225 9378 196185Number of countries 119 119 13 119HS six-digit fixed effects No No No YesCountry fixed effects Yes Yes Yes No

OLS estimates. Robust Standard errors in parenthesis: ∗∗ and ∗ stand for 5 % and 10 % statistical signifi-

cance.

Table 4: Is market power used within tariff waters? OLS estimates

(1) (2) (3) (4) (5) (6)

Import Market Power -6.88e-07(1/e∗) ( 1.39e-05 )

Log of Import market Power -0.0019∗∗

(log(1/e∗)) (1.37e-04)

Dummy for High and Medium Power -0.0047∗∗

(within each country) (0.0007)

Dummy for High Power -0.007∗∗

(within each country) (0.001)

Dummy for Medium Power -0.0028∗∗

(within each country) (0.0003)

Dummy for High and Medium Power -0.0071∗∗

(across all countries) (0.0009)

Dummy for High Power -0.0129∗∗

(across all countries) (0.0018)

Dummy for Medium Power -0.0049∗∗

(Med in all sample) (0.0004)

Water -0.062∗∗ -0.0481∗∗ -0.0749∗∗ -0.0753∗∗ -0.0777∗∗ -0.0789∗∗

(0.0097) (0.0132) (0.0063) (0.0061) (0.0061) (0.0057)

Power∗water -3.4e-05 0.0047∗∗ 0.0182∗∗ 0.0311∗∗ 0.0239∗∗ 0.0443∗∗

(High in columns 4 and 6) (5.86e-05) (0.001) (0.0051) (0.0104) (0.0061) (0.0129)

Medium market Power∗water 0.0060∗∗ 0.0073∗∗

(Medium in columns 4 and 6) (0.0015) (0.0016)

Uses power when water is ≥ -2.02p.p. 40.42p.p.∗∗ 25.82p.p.∗∗ 19.35p.p.∗∗ 29.71p.p∗∗ 29.12p.p.∗∗

(High power in columns 4 and 6) (2.06) (7.09) (3921) (2.37) (4.35) (4.60)

Uses power when water is ≥ 46.67p.p.∗∗ 67.12p.p∗∗

(Medium in columns 4 and 6) (8.52) (11.45)

OLS estimates. All columns include year, HS six-digit and country x year x HS two-digit fixed effects. Robust standard errors in parenthesis:

∗∗ and ∗ stand for 5 % and 10 % statistical significance. F-statistics indicate all regressions are significant at the 1 percent level. Number of

observations in each specification is 1690909.

Table 5: Is market power used within tariff waters? IV estimates

(1) (2) (3) (4) (5) (6)

Import Market Power -0.0159∗∗

(1/e∗) (0.0033)

Log of Import market Power -0.0168∗∗

(log(1/e∗)) (0.0016)

Dummy for High and Medium Power -0.0026(within each country) (0.0042)

Dummy for High Power -0.0083∗

(within each country) (0.0046)

Dummy for Medium Power -0.0871∗∗

(within each country) (0.0226)

Dummy for High and Medium Power -0.0821∗∗

(across all countries) (0.0049)

Dummy for High Power -0.059∗∗

(across all countries) (0.0191)

Dummy for Medium Power -0.0285(Med in all sample) (0.0439)

Water -0.0614∗∗ -0.6197∗ -0.1005∗∗ -0.129∗∗ -0.153∗∗ -0.1299∗∗

(0.0064) (0.3611) (0.0142) (0.0165) (0.0138) (0.0267)

Power∗water -0.0099 -0.1616 0.0630∗∗ 0.0902∗∗ 0.1336∗∗ 0.1034∗∗

(High in columns 4 and 6) (0.0294) (0.1062) (0.0232) (0.0268) (0.0207) (0.0258)

Medium market Power∗water 0.1031∗∗ 0.1071∗∗

(Medium in columns 4 and 6) (0.0309) (0.0535)

Uses power when water is ≥ -160.61p.p. -10.39p.p. 0 p.p 9.09p.p.∗∗ 61.45p.p∗∗ 57.06p.p.∗∗

(High power in columns 4 and 6) (484.25) (7.66) (5.95) (4.02) (8.21) (12.07)

Uses power when water is ≥ 84.48p.p.∗∗

26.61p.p.(Medium in columns 4 and 6) (23.4) (31.14)

Hansen’s Orthogonality Test 6.26 0.842 0.499 1.49 0.050 4.12(p-value) (0.01) (0.36) (0.48) (0.47) (0.823) (0.13)

Kleibergen-Paap’s Weak IV Test 1.790 0.290 471.37 10.38 686.10 4.62(pass 5 percent critical value?) N N Y Y Y Y

GMM variable estimates. All columns include year, HS six-digit and country x year x HS two-digit fixed effects. Standard errors in paren-

thesis: ∗∗ and ∗ stand for 5 % and 10 % statistical significance. Instruments for water and power include: water vapour, the average import

demand elasticity in the rest-of-the-world for a given HS six-digit good and country, the interaction between water vapour and the average import

demand elasticity in the rest-of-the-world, and between water vapour and the average across countries of the export supply elasticity from the

point of view of exporters. Columns 3 to 6 use dummies derived from these variables as in Broda et al. (2008). High correspond to the top third

of the distribution and Medium to the those in the middle of the distribution. Columns 3 and 4 calculate these dummies within each country

elasticity distribution, whereas columns 5 and 6 calculate these dummies across all countries. F-statistics are not displayed but they suggest that

all estimated models are statistically significant at the 1 percent level. Number of observations in each specification is 1562047.

Table 6: Market Power and Fear of Retaliation – IV estimates

(1) - low fear (2) - high fear (3) - low fear (4) - high fear

Log of Import market Power -0.0087∗∗ -0.0064∗∗

(log(1/e∗)) (0.0015) (0.0025)

Dummy for High and Medium Power -0.0153 -0.0742∗∗

(across all countries) (0.0102) (0.0078)

Water 0.2303 -0.3166∗∗ -0.0814∗∗ -0.139∗∗

(0.1667) (0.1329) (0.014) (0.0362)

Power∗water 0.0779∗ -0.0744∗ 0.0273∗ 0.0763(High in columns 4 and 6) (0.042) (0.042) (0.0151) (0.0525)

Hansen’s Orthogonality Test 2.585 7.231 0.255 25.23(p-value) (0.11) (0.01) (0.61) (0.00)

Kleibergen-Paap’s Weak IV Test 1.118 1.069 98.94 465.95(pass 5 percent critical value?) N N Y Y

GMM variable estimates. All columns include year, HS six-digit and country x year x HS two-digit fixed effects. Standard errors in parenthe-

sis: ∗∗ and ∗ stand for 5 % and 10 % statistical significance. Instruments for water and power include: water vapour, the average import demand

elasticity in the rest-of-the-world for a given HS six-digit good and country, the interaction between water vapour and the average import demand

elasticity in the rest-of-the-world, and between water vapour and the average across countries of the export supply elasticity from the point of view

of exporters. F-statistics are not displayed but they suggest that all estimated models are statistically significant at the 1 percent level. Number

of observations in columns 1 and 2 is 429469 and 557664, respectively. In the case of columns 3 and 4, the number of observations is 358669 and

675740, respectively.

Figure 1Distribution of the inverse of export supply elasticities faced by importers

41

Figure 2Correlation between the export supply elasticities faced by importers

and those calculated using equation (25)